An NVIDIA Tensor core GPU with 54 billion transistors. Introduced in 2020 and using the Ampere architecture, the A100 was NVIDIA's top-of-the-line Tensor core GPU prior to the H100.
Lots of A100s
To train AI large language models with a trillion or more parameters (data examples), datacenters need several thousand A100s, each costing in the neighborhood of $10,000. See
Tensor core and
H100.